Most people with arthritis are not regularly active. Understanding what factors influence exercise is essential for designing programs to increase participation. The objective of this study was to examine the correlates of exercise in people with arthritis. Using a cross-sectional design, sociodemographic, health-related, and psychosocial variables were collected from community-dwelling individuals with arthritis (N = 141). Associations with exercise level were examined with bivariate statistics (ANOVAs, chi-squares) and logistic-regression analyses. Exercisers were less likely than nonexercisers and insufficiently active people to report that arthritis negatively affected their physical and social functioning, and they reported more positive affect and greater self-efficacy (p < .05). Exercisers also reported less pain than nonexercisers (p < .05). In multiple logistic-regression analyses, self-efficacy and physical limitations remained independent predictors of exercise. The results suggest the need to target exercise self-efficacy when designing exercise interventions. Results also suggest the need to tailor exercise programs to individuals' physical limitations.